Development of an AI-Assisted Diagnostic Tool for Mycosis Fungoides and Other Cutaneous Lymphoproliferative Diseases Using Microscopic Image Analysis: A Training and Validation Study

Cutaneous lymphoproliferative diseases (CLPDs) encompass both benign and malignant disorders, ranging from pseudolymphomas to cutaneous T-cell and B-cell lymphomas. Mycosis fungoides (MF) is the most common malignant subtype, but diagnosis is often challenging because benign inflammatory conditions can closely mimic lymphoma. Histopathological examination remains the gold standard, yet interpretation is subjective and prone to inter-observer variability. This highlights the need for standardized diagnostic tools, including artificial intelligence (AI) solutions.

This study is a retrospective diagnostic accuracy investigation using routinely collected data. Archived hematoxylin and eosin (H&E) stained slides of patients with MF and other CLPDs will be retrieved from the Dermatopathology Unit at Kasr Al-Aini Hospitals, Cairo University. Slides of benign mimickers such as pseudolymphoma and pityriasis lichenoides will also be included. Cases with poor slide quality or insufficient data will be excluded.

Digitized images will be captured using both high-resolution microscope cameras and standardized smartphone devices to evaluate feasibility. Experienced dermatopathologists will annotate regions of interest, and relevant clinical data will be extracted to build a structured database. Deep learning models, particularly convolutional neural networks (CNNs), will be trained and validated on these datasets. Preprocessing techniques such as color normalization, stain separation, and data augmentation will be applied to enhance robustness.

The primary outcomes are diagnostic accuracy, sensitivity, specificity, and predictive values of the AI models in differentiating malignant from benign CLPDs, and in staging MF. Secondary outcomes include comparison with expert dermatopathologists, assessment of smartphone-based imaging, and evaluation across magnification levels. More than 500 slides collected over the past five years will be used, divided into training, validation, and testing sets.

By integrating AI into dermatopathology, this study aims to reduce diagnostic variability, improve accuracy, and explore novel imaging approaches. The work represents one of the first applications of deep learning to CLPDs, with potential to enhance patient care in both specialized centers and resource-limited healthcare settings.

Inclusion criteria

• Archived slides of patients with a confirmed histopathological diagnosis of malignant CLPDs (e.g., mycosis fungoides at all stages, cutaneous B-cell lymphoma, primary cutaneous anaplastic large cell lymphoma, lymphomatoid papulosis), based on WHO-EORTC criteria.
• Archived slides of patients with benign CLPDs that mimic MF clinically and histologically (e.g., pseudolymphoma, pityriasis lichenoides chronica, pityriasis lichenoides et varioliformis acuta [PLEVA]).
• Availability of adequate quality hematoxylin and eosin (H&E) stained slides.
• Availability of relevant clinical data (age, sex, disease duration, distribution of lesions, drug history).

Exclusion criteria

• Slides with significant artifacts (folding, tearing, poor staining) that prevent adequate image analysis.
• Cases with insufficient clinical or pathological data for definitive diagnosis.
• Cases with secondary cutaneous CLPDs